Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
anthropic-pptx
v1.0.0Use this skill any time a .pptx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations;...
⭐ 0· 18·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
high confidencePurpose & Capability
The name/description (pptx handling) matches the included Python scripts and documentation: unpack, edit, thumbnail, and pack workflows for .pptx files. However, the LICENSE.txt claims Anthropic proprietary materials while the skill source/owner is unknown — a licensing/attribution mismatch that users should notice.
Instruction Scope
SKILL.md instructs running Python commands and the included scripts (thumbnail.py, unpack/pack/clean). Those scripts call external tools (soffice, pdftoppm) and expect Python modules (Pillow, defusedxml, etc.) that are not declared. The docs also instruct using subagents and an 'Edit tool' which could broaden file access. The guidance triggers whenever the user mentions slides/decks, increasing the chance of frequent invocation. The instructions therefore access system commands and files beyond what SKILL.md declares.
Install Mechanism
There is no install spec even though the skill includes many code files and XML schemas. At runtime scripts/soffice.py will write C source into a tempdir and invoke gcc to compile a shared object, then set LD_PRELOAD for subprocesses. That runtime compilation and LD_PRELOAD behavior is not called out in SKILL.md and is higher-risk than ordinary pure-Python scripts.
Credentials
The skill does not request credentials or environment variables in its metadata. The code copies os.environ and may set SAL_USE_VCLPLUGIN and LD_PRELOAD for soffice subprocesses, but it does not attempt to read or exfiltrate secrets. Still, the use of LD_PRELOAD and copying the environment is an environment-modifying action that should be acknowledged.
Persistence & Privilege
always:false and no skills/config modification are requested. However, the skill writes files (temp C source and a compiled .so) and may set LD_PRELOAD for subprocesses at runtime — this is not persistent across agent installs but does give the skill the ability to alter subprocess behavior on the host during use. That elevated runtime capability increases the attack surface compared with a pure-Python library.
What to consider before installing
What to consider before installing:
- Missing dependency declarations: the skill runs soffice, pdftoppm, gcc and requires Python packages (Pillow, defusedxml, etc.) but the metadata lists no required binaries or packages. If you install this, ensure those tools are available and understand they will be invoked.
- Runtime compilation + LD_PRELOAD: the included office/soffice.py will write C source to a temp directory and compile a shared object with gcc, then set LD_PRELOAD for LibreOffice subprocesses. That changes how subprocess socket calls behave while the shim is loaded. Only run this skill in an environment you control (container or VM) and if you trust the code.
- Licensing/attribution mismatch: LICENSE.txt claims Anthropic ownership but the source/owner is unknown. Confirm you have the right to use these materials and that the license terms are acceptable for your use.
- Frequent/autonomous triggers: SKILL.md tells agents to trigger this skill whenever slides/presentations are mentioned. If the agent can invoke skills autonomously, this may cause the skill to run often; consider limiting invocation scope or requiring explicit user approval.
- Audit the code and run in isolation: review the provided scripts (they are readable) and test them in an isolated environment before giving the skill access to sensitive files. If you do not want on-host compilation or LD_PRELOAD behavior, remove or modify the soffice shim and ensure the environment has a normal soffice/pdftoppm setup.
If you trust the author and will run in a controlled environment (or update the metadata to declare dependencies), the skill appears coherent with its stated purpose; otherwise treat it cautiously.Like a lobster shell, security has layers — review code before you run it.
latestvk979dd7ns7pzakpzc2j5699fe1851ehj
License
MIT-0
Free to use, modify, and redistribute. No attribution required.